Aiming at the problems of rapidly-exploring random tree (RRT) algorithm in obstacle avoidance path planning, such as weak adaptability to maps, poor sampling quality, many invalid nodes, long planning time and poor path quality, an improved RRT algorithm was proposed. Firstly, on the basis of the traditional RRT algorithm, the map complexity evaluation strategy was used to calculate the appropriate step size and bias probability, so as to realize the self-adaptation to different maps. Then, through the dynamic update strategy of sampling area, the random tree was sampled in the effective area to ensure the positive growth. After the sampling area was determined, the sampling point optimization strategy was adopted to improve the effectiveness of sampling points and make the random tree grow near the target points. Finally, the node reconnection strategy was used to optimize the planned initial obstacle avoidance path, and an obstacle avoidance path with fewer bending times was obtained. The feasibility of the improved RRT algorithm was verified in Python and MATLAB environments. The results showed that the improved RRT algorithm could quickly plan a collision-free high-quality path for maps with different complexity and when applied to robotic arms. The research results can provide reference for improving efficiency of robot obstacle avoidance path planning.